On the Utility of Big Data in Predicting Short-term Interstitial/Blood Glucose Levels in Type 2 Diabetes Mellitus through Machine Learning Techniques
|Ioannis Chatzigiannakis||Componenti strutturati del gruppo di ricerca|
Data have shown that continuous glucose monitoring (CGM) in patients with type 2 diabetes (T2D) can not only help achieve HbA1c improvements but also identify unforeseen hypoglycemia risk and address glucose variability. CGM in clinical practice can be deployed both in the professional, clinical setting (retrospective review) and being used in the personal, at-home setting (real time). RT-CGM, flash CGM, or blinded CGM (no data displayed to the patient) can be worn for 3, 7, or 14 days and then downloaded in the clinic for interpretation. The current recommendation for accurate or reproducible pattern recognition is to analyze 14 days of CGM data. The data can be evaluated by the clinician, compared to the last CGM profile done and reviewed with the patient to help drive treatment changes and/or improve patient¿s self-management skills. Meanwhile, numerous wearable devices -now linked to the CGM devices- have been introduced allowing to monitor different physiological aspects of diabetic people such as exercise & activity, heart rate & electrocardiogram, eating behaviours, thus completing the characterization of a diabetic person life and generating a really huge amount of data. Such data could potentially help in better predicting the magnitude, tendency, frequency and duration of fluctuations of glucose levels. Along with the aforementioned benefits, the use of multiple wearable devices introduces the necessity of their daily maintenance which could potentially lead to a decline of the perceived quality of life. It is therefore important to understand the benefits of creating big data sets, in relation to the greater computational efforts required to process the data, the introduced overhead to the users for collecting the data, and of course, the repercussion of the misuse of the data. In this project we would like to focus on how far we can get in predicting interstitial glucose levels (IGL) to regulate metabolic control without using large volumes of data.